Anthropic and Amazon are going all in on compute: $100B+ and 5GW of new capacity

Anthropic just dropped a bombshell of an announcement that makes their previous partnership with Amazon look like a warm-up act. They’ve signed a new agreement that commits over $100 billion to AWS technologies over the next ten years, securing up to 5 gigawatts of compute capacity for training and running Claude. That’s not a typo — 5 gigawatts is enough to power a small city, and it’s all going into AI infrastructure.

Let me break down what’s actually new here, because the press release is dense.

The numbers that matter

First, the immediate stuff: significant Trainium2 capacity is coming online in Q2 of this year (we’re in April 2026 as I write this), and nearly 1 gigawatt total of Trainium2 and Trainium3 capacity will be online by the end of 2026. That’s not theoretical future planning — that’s compute arriving in the next three months.

The long-term commitment spans Graviton and Trainium chips from generation 2 through 4, with the option to buy future generations as they become available. Amazon is also investing $5 billion in Anthropic today, with potential for up to an additional $20 billion down the road. That’s on top of the $8 billion Amazon has already put in.

Why this matters right now

Anthropic’s run-rate revenue has hit $30 billion, up from about $9 billion at the end of 2025. That’s 3x growth in roughly 15 months. And they’re honest about the pain: “our unprecedented consumer growth, in particular, has impacted reliability and performance for free, Pro, Max, and Team users, especially during peak hours.” Translation: Claude is getting hammered by demand, and they need more compute fast.

This deal is the answer. They’re getting meaningful compute in the next three months, with nearly 1GW total before year-end.

What changes for users

The full Claude Platform will be available directly within AWS — same account, same controls, same billing, no extra credentials or contracts. That’s a big deal for enterprise customers who need to stay within their existing governance frameworks. Claude remains the only frontier AI model available on all three major clouds: AWS Bedrock, Google Cloud Vertex AI, and Microsoft Azure Foundry. That’s a nice flex.

They’re also expanding inference capacity in Asia and Europe, which should help with latency for international users.

My take

This level of infrastructure commitment is unprecedented. $100 billion over ten years for compute is not something you see every day. It signals that Anthropic is betting everything on continued exponential growth in demand for Claude, and Amazon is betting that custom silicon (Trainium) is the right horse to ride.

The $30 billion revenue run rate is the number that caught my eye. That’s serious money, and it explains why Amazon is willing to keep writing checks. But I also notice the honesty about reliability problems — consumer growth is straining the system. If they can’t keep Claude responsive during peak hours, no amount of infrastructure spending will matter if users get frustrated and leave.

Also worth noting: they’re using over one million Trainium2 chips right now to train and serve Claude. That’s a massive cluster, and they’re about to scale it dramatically.

The bottom line

This is Anthropic and Amazon doubling down on each other in a way that makes the partnership effectively exclusive for training infrastructure. Anthropic gets the compute it desperately needs, Amazon gets a flagship customer for its custom silicon. The next 12 months will tell us whether this kind of vertical integration pays off or if it creates lock-in risks. My bet is on the former, but I’ve been wrong before.